[GALLERY] The super-cheap Claude AI API turns out to be selling user data.

The Claude AI API black market in China is selling access for up to 90% less, but behind the attractive price lies the risk of exposing all personal and work data.

An unusually low-cost AI service is booming in China, with numerous accounts on Taobao, Telegram, GitHub, and programming forums simultaneously selling access to Anthropic Claude AI at prices up to 90% lower than the official API, creating an "underground market" that is raising alarms in the global tech community.

Many vendors advertise that for a fraction of the cost, users can access unlimited Claude Opus, GPT-5, or Gemini, supporting both Cursor and VSCode without a VPN, while promising response speeds comparable to native servers from OpenAI or Anthropic.

According to an investigation by the Oxford China Policy Lab, these services operate through a network of "transfer stations"—AI relay stations with servers located overseas that forward user requests to large AI models and then return the results, overcoming access barriers and cutting costs to extremely low levels.

Alarmingly, the vast majority of these cheap API accounts are created by exploiting free credit, illegally sharing business packages, or even using stolen credit card information, while some networks even hire people in low-income countries to verify identities with paper documents and facial scans to bypass Anthropic's censorship.

However, the biggest shock was that users didn't receive the AI ​​model they thought they were paying for, as research from the CISPA Helmholtz Center showed that many proxy services swapped models, for example, advertising Gemini-2.5 but actually returning weaker models, or counterfeit Chinese domestic AI like Qwen and MiniMax.

Even more dangerously, all commands, source code, internal business data, and AI responses are recorded on the intermediary's server, turning users into "living data mines" for information-mining networks with virtually no privacy protection mechanisms.

Experts warn that this poses a particularly serious risk to programmers and businesses, as many AI coding agents are now used to process source code, validation logic, and internal APIs. This means that using the black market Claude AI service could result in all company data being transferred directly to an unknown third-party server.

The battle between American AI companies and the shadow proxy market is escalating sharply, with Anthropic reporting the discovery of approximately 24,000 fraudulent accounts linked to numerous Chinese AI companies. This reveals that behind the advertisements for "cheap AI" lies an industrial-scale data mining ecosystem where users believe they are getting a bargain but end up being sold off.

Source: https://khoahocdoisong.vn/gallery-api-claude-ai-sieu-re-hoa-ra-dang-ban-luon-du-lieu-nguoi-dung-post2149100859.html